This is not the erudite, entertaining and incisively written diary#

Tue, 13 May 2003 18:19:58 +0000

This is not the erudite, entertaining and incisively written diary entry that you were doubtless (and foolishly) hoping for after more than a week's lapse in entries. It's still on my TODO list, but the date has been pushed back to late 2007

This is basically just a note to point out that Kevin Rosenberg is organising an AMD64 (formerly x86-64, formerly Hammer, née Sledgehammer) port of SBCL. If you're interested in helping fund development on this, have a look at the page on pubsoft. We're already, ooh, 1/40th of the way to the target. And it's tax-deductable, if you're a US taxpayer.

Declaration of interest: I'm bidding (or more accurately, metacircles is bidding) on the project.

(Volunteers to produce a cute thermometer graphic a la "church roof appeal", contact Kevin or me ;-)

The TODO list is looking entirely prosaic right now:#

Sat, 17 May 2003 04:24:23 +0000

The TODO list is looking entirely prosaic right now:

• article for Linux User magazine on something to do with CLiki (needs finishing)

• paper for UKUUG conference (needs starting)

• book proposal, revisions to

• try to work out if I want to/have the time to/can afford to go to Metz for the LSM2003, and if so how much of the UKUUG paper I can recycle and how much I need to write fresh.

So obviously this is a good time to spend a week looking at SBCL 64 bit issues. In fact I can't think of a more productive use of time, provided I discount almost anything else involving a computer in some way. As of my most recent checkin, though, we're at the point where

• the runtime support builds
• all the Lisp files build in the crosscompiler
• the initial image can be written out
• the initial image is loaded into the runtime, and starts executing. Even calls functions and stuff
• but dies fairly quickly when we start doing floating point

<dan_b> 0x00000000321b927c in ?? ()
<dan_b> 0x321B8FE0: MAKE-HASH-TABLE   #X321B8FC9
<dan_b> 0x321BB140: HASH-TABLE-COUNT   #X321BB129
<Krystof> Cooool
<dan_b> this brings back memories
<Krystof> yeah, we have been here(tm)

The particular memories I have in mind are from summer/autumn 2000 when I first started the SBCL Alpha port, and I think I was stuck with a problem in this area for some few months. I expect this time it'll be a matter of days or possibly even hours; what's frightening here is the realisation of how long ago those days were and how long I must have spent reading SBCL code since then. If I'd devoted the same amount of time to gcc, I'd probably be rich by now. Or, anyway, richer.

Nobody reported any interesting bugs in the sbcl prerelease last month, so we'll probably go ahead and do a 0.8 this month. Which means code freeze in about three days time, which means it's time to turn my attention to some threading cleanups this weekend. Definitely time to stop messing around with 64 bit support.

Lately I've been rereading#

Sat, 24 May 2003 14:39:02 +0000

Lately I've been rereading

• The Manual (How to have a Number One - the Easy Way). Summary: doesn't sound that easy, or, to be honest, a whole lot of fun. Good reading, though

  Firstly, it has to have a dance groove that will run all the way through the record and that the current 7" buying generation will find irresistible. Secondly, it must be no longer than three minutes and thirty seconds (just under 3'20 is preferable). If they are any longer Radio One daytime DJs will start fading early or talking over the end, when the chorus is finally being hammered home - the most important part of any record. Thirdly, it must consist of an intro, a verse, a chorus, second verse, a second chorus, a breakdown section, back into a double length chorus and outro. Fourthly, lyrics. You will need some, but not many.

  (The 'next'/'previous' links on that document are broken, yes. You have to edit the URL in the title bar to read beyond the first chapter)

• Black Vinyl, White Powder, although by dipping into it rather than reading through in sequence: some really interesting subject matter, which is lucky because most of the writing itself falls absolutely flat - with few exceptions

  Cliff's backing group, the Shadows, also became famous. Their speciality was to perform while walking with strange criss-cross steps as if they were trying to untangle their feet from a mike cable. Their influence was enormous. Within a year, every 16-year-old in Britain was capable of untangling his feet from a mike cable while playing the guitar and looking straight in front of him

In recent months, I've become teed off enough with the speed of#

Tue, 27 May 2003 02:43:28 +0000

In recent months, I've become teed off enough with the speed of ifile-gnus that I find myself doing all kinds of stuff to avoid actually pulling my mail into gnus, often for days at a time. Typically this means running mutt on my shell host, which means that my mood for the day is set by checking email in the morning and deleting 100 assorted get rich quick schemes, offers of pharmaceuticals, viruses and invitations to pornographic web sites which are probably illegal and damn well should be. This may be one of the reasons I've been so cheerful and enthusiastic lately, and ought to be fixable.

The choices open to me, it seems: (a) hack up some kind of persistent ifile interface (yay, another network daemon running on the loopback interface), (b) installing some other gnus-based filter, or (c) saying "sod it, I'll go back to spamassassin".

Someone on the Gnus team seems to have taken the unix mechanism-not-policy thing to heart. Here's a tip: just because you have the full power of Lisp as your configuration language is no reason to make people have to read your entire source code just to find out how to make your program work at all. If it's actually possible to use spam-stat to classify incoming mail into different folders for spam vs non-spam, it's seriously non-obvious from the info file. Spam filtering is, unfortunately, a basic requirement of a mail reader these days: a mail reader that doesn't do it out of the box will soon be about as useful as a mail reader that doesn't use a pager of some kind to display messages a screen at a time. Still, this is CVS gnus so carping about the intallation is perhaps unfair if they haven't had time to work on it yet.

So, instead, I've reverted to spamassassin/spamd. It's just about usable after setting OPTIONS="-c -m 1" in /etc/default/spamassassin (though putting 600 messages through it using formail it still managed to push the loadavg to 63 briefly and then give me "Can't fork" messages, so I don't think that's the whole answer). It still does a worse job of identifying spam based on looking at the entire message body than I can based on the header alone. And for each incoming message I have inetd spawn sendmail (actually Exim) spawn sh spawn spamc | procmail. So, Exim reads the message from the network and outputs it to spamc. spamc reads it and outputs it to spamd. spamd reads it and outputs it, suitably mutilated, to spamc. spamc reads it and outputs it to procmail. procmail reads it and outputs it to disk. And this is the normal way of arranging things, if the documentation is to be believed.

I read, sometimes, in LWN, about the marvellous things the kernel developers are doing to reduce the context switch time or process startup time, or the amount of copying done to move packets around the network stack, and I can only conclude that it's all a complete and utter waste of time when this is the kind of crap that people build on top of it.

Still, if it means I can start my working day tomorrow without ten messages from someone eager to tell me that he/she "jammed the babysitter" (what is this, CB radio?) maybe it's all worthwhile.

Roll on pervasive PGP, really. If you should ever bump into me physically, ask for my key fingerprint, because I am -><- this close to whitelisting signed mail from real people and letting everything else queue up in the "Things to do - urgent" folder.

Installed Firebird 0.6, and it seems to have fixed the weird problem with#

Tue, 27 May 2003 20:16:26 +0000

Installed Firebird 0.6, and it seems to have fixed the weird problem with hyphens and apostrophes that (on some but not all web pages) would cause 0.5'
irritating, but I never found out whether it was a phoenix problem, a mozilla problem, a debian packaging problem, a font problem, a locale problem, or frankly, what. Which is why, if anyone associated with the mozilla project is reading this, i didn't report it to them. Anyway, if anyone else on the planet was suffering that one, 0.6 is the cure. Now I can read Mike Shaver's diary again.